dulwich-users team mailing list archive

[Jelmer Vernooij <jelmer@xxxxxxxxx>]

On Wed, Oct 16, 2013 at 10:46:10AM +0800, Crispin Wellington wrote:
> OK. digging a bit further it seems that the push is actually 'working'. If
> I don't make any commit, just pull and then push, I get the aforementioned
> error. I suppose there is nothing to send, but dulwich advertises that
> there is something to send and then sends nothing causing an remote
> unexpected eof error? perhaps...
What version of dulwich are you using? There were some fixes for pushes
in 0.9.1.

> If I pull, change a file, and commit it. Then push. I also get an error.
> But the commit does get pushed and appears on the server. With a commit I
> get "UpdateRefsError: HEAD, refs/heads/develop failed to update".

> It seems the remote server has trouble updating its refs? If I pull the
> repo fresh (using the same pull code)  I get the repo BEFORE the commit. If
> I then issue another change and commit, and push, I again get the
> UpdateRefsError, and on bitbucket, the old commit has now dissapeared and
> been REPLACED by the new one!!!
> 
> It's like the remote is accepting the blob, adding it to the repository,
> but not updating its refs to point at that new HEAD.
> 
> debugging in as before to this "UpdateRefsError" I get the following
> response:
> 
> ipdb> resp.code
> 200
> ipdb> resp.msg
> 'OK'
> ipdb> resp.url
> 'https://bitbucket.org/cwellington/raven.git/git-receive-pack'
> ipdb> resp.read()
> "0088\x02error: refusing inconsistent update between symref 'HEAD'
> (0000000..1ff6474) and its target 'refs/heads/master'
> (c7c2899..1ff6474)\n0074\x01000eunpack ok\n0028ng HEAD inconsistent aliased
> update\n0035ng refs/heads/master inconsistent aliased update\n00000000"
> 
> Any ideas?
It looks like HEAD in the remote repo is a symref to refs/heads/master.
I think it's telling you that if you update one to a particular ref, you should
update the other to that particular ref too.

> On Wed, Oct 16, 2013 at 10:01 AM, Crispin Wellington <crispinw@xxxxxxxxxx>wrote:
> 
> > So I put the debugger in to break at dulwich/client.py line 924 ish...
> > just after 'resp' response object is created. Seems to be the remote server
> > telling me it has had problems unpacking HEAD:
> >
> > > /.../python2.7/site-packages/dulwich/client.py(926)send_pack()
> >     925
> > --> 926         resp_proto = Protocol(resp.read, None)
> >     927         self._handle_receive_pack_tail(resp_proto,
> > negotiated_capabilities,
> >
> > ipdb> resp
> > <addinfourl at 30480288 whose fp = <socket._fileobject object at
> > 0x1ce5650>>
> > ipdb> resp.code
> > 200
> > ipdb> resp.msg
> > 'OK'
> > ipdb> resp.url
> > 'https://bitbucket.org/username/project.git/git-receive-pack'
> > ipdb> resp.read()
> > '00a6\x010031unpack eof before pack header was fully read\n001bng HEAD
> > unpacker error\n0029ng refs/heads/develop unpacker error\n0028ng
> > refs/heads/master unpacker error\n00000000'
> > ipdb>
> >
> > I'm not familiar with the git http(s) 'dumb' protocol. Is it documented
> > anywhere?
You're using the smart protocol, not the dumb protocol. The latter just 
accesses a raw git repository using plain HTTP GET requests.

Cheers,

Jelmer